Use signalfd(2) for epoll latches.

macdice · macdice · commit 6a2a70a02018 · 2021-03-01T14:12:02.000+13:00
Cut down on system calls and other overheads by reading from a signalfd instead of using a signal handler and self-pipe. Affects Linux sytems, and possibly others including illumos that implement the Linux epoll and signalfd interfaces. Reviewed-by: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CA+hUKGJjxPDpzBE0a3hyUywBvaZuC89yx3jK9RFZgfv_KHU7gg@mail.gmail.com
diff --git a/src/backend/libpq/pqsignal.c b/src/backend/libpq/pqsignal.c
@@ -35,13 +35,15 @@ sigset_t	UnBlockSig,
  * collection; it's essentially BlockSig minus SIGTERM, SIGQUIT, SIGALRM.
  *
  * UnBlockSig is the set of signals to block when we don't want to block
- * signals (is this ever nonzero??)
+ * signals.
  */
 void
 pqinitmask(void)
 {
 	sigemptyset(&UnBlockSig);
 
+	/* Note: InitializeLatchSupport() modifies UnBlockSig. */
+
 	/* First set all signals, then clear some. */
 	sigfillset(&BlockSig);
 	sigfillset(&StartupBlockSig);
diff --git a/src/backend/storage/ipc/latch.c b/src/backend/storage/ipc/latch.c
@@ -3,21 +3,20 @@
  * latch.c
  *	  Routines for inter-process latches
  *
- * The Unix implementation uses the so-called self-pipe trick to overcome the
- * race condition involved with poll() (or epoll_wait() on linux) and setting
- * a global flag in the signal handler. When a latch is set and the current
- * process is waiting for it, the signal handler wakes up the poll() in
- * WaitLatch by writing a byte to a pipe. A signal by itself doesn't interrupt
- * poll() on all platforms, and even on platforms where it does, a signal that
- * arrives just before the poll() call does not prevent poll() from entering
- * sleep. An incoming byte on a pipe however reliably interrupts the sleep,
- * and causes poll() to return immediately even if the signal arrives before
- * poll() begins.
+ * The poll() implementation uses the so-called self-pipe trick to overcome the
+ * race condition involved with poll() and setting a global flag in the signal
+ * handler. When a latch is set and the current process is waiting for it, the
+ * signal handler wakes up the poll() in WaitLatch by writing a byte to a pipe.
+ * A signal by itself doesn't interrupt poll() on all platforms, and even on
+ * platforms where it does, a signal that arrives just before the poll() call
+ * does not prevent poll() from entering sleep. An incoming byte on a pipe
+ * however reliably interrupts the sleep, and causes poll() to return
+ * immediately even if the signal arrives before poll() begins.
  *
- * When SetLatch is called from the same process that owns the latch,
- * SetLatch writes the byte directly to the pipe. If it's owned by another
- * process, SIGURG is sent and the signal handler in the waiting process
- * writes the byte to the pipe on behalf of the signaling process.
+ * The epoll() implementation overcomes the race with a different technique: it
+ * keeps SIGURG blocked and consumes from a signalfd() descriptor instead.  We
+ * don't need to register a signal handler or create our own self-pipe.  We
+ * assume that any system that has Linux epoll() also has Linux signalfd().
  *
  * The Windows implementation uses Windows events that are inherited by all
  * postmaster child processes. There's no need for the self-pipe trick there.
@@ -46,6 +45,7 @@
 #include <poll.h>
 #endif
 
+#include "libpq/pqsignal.h"
 #include "miscadmin.h"
 #include "pgstat.h"
 #include "port/atomics.h"
@@ -79,6 +79,10 @@
 #error "no wait set implementation available"
 #endif
 
+#ifdef WAIT_USE_EPOLL
+#include <sys/signalfd.h>
+#endif
+
 /* typedef in latch.h */
 struct WaitEventSet
 {
@@ -139,7 +143,14 @@ static WaitEventSet *LatchWaitSet;
 #ifndef WIN32
 /* Are we currently in WaitLatch? The signal handler would like to know. */
 static volatile sig_atomic_t waiting = false;
+#endif
 
+#ifdef WAIT_USE_EPOLL
+/* On Linux, we'll receive SIGURG via a signalfd file descriptor. */
+static int	signal_fd = -1;
+#endif
+
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 /* Read and write ends of the self-pipe */
 static int	selfpipe_readfd = -1;
 static int	selfpipe_writefd = -1;
@@ -150,8 +161,11 @@ static int	selfpipe_owner_pid = 0;
 /* Private function prototypes */
 static void latch_sigurg_handler(SIGNAL_ARGS);
 static void sendSelfPipeByte(void);
-static void drainSelfPipe(void);
-#endif							/* WIN32 */
+#endif
+
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_EPOLL)
+static void drain(void);
+#endif
 
 #if defined(WAIT_USE_EPOLL)
 static void WaitEventAdjustEpoll(WaitEventSet *set, WaitEvent *event, int action);
@@ -175,7 +189,7 @@ static inline int WaitEventSetWaitBlock(WaitEventSet *set, int cur_timeout,
 void
 InitializeLatchSupport(void)
 {
-#ifndef WIN32
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 	int			pipefd[2];
 
 	if (IsUnderPostmaster)
@@ -247,8 +261,21 @@ InitializeLatchSupport(void)
 	ReserveExternalFD();
 
 	pqsignal(SIGURG, latch_sigurg_handler);
-#else
-	/* currently, nothing to do here for Windows */
+#endif
+
+#ifdef WAIT_USE_EPOLL
+	sigset_t	signalfd_mask;
+
+	/* Block SIGURG, because we'll receive it through a signalfd. */
+	sigaddset(&UnBlockSig, SIGURG);
+
+	/* Set up the signalfd to receive SIGURG notifications. */
+	sigemptyset(&signalfd_mask);
+	sigaddset(&signalfd_mask, SIGURG);
+	signal_fd = signalfd(-1, &signalfd_mask, SFD_NONBLOCK | SFD_CLOEXEC);
+	if (signal_fd < 0)
+		elog(FATAL, "signalfd() failed");
+	ReserveExternalFD();
 #endif
 }
 
@@ -273,19 +300,28 @@ InitializeLatchWaitSet(void)
 void
 ShutdownLatchSupport(void)
 {
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 	pqsignal(SIGURG, SIG_IGN);
+#endif
 
 	if (LatchWaitSet)
 	{
 		FreeWaitEventSet(LatchWaitSet);
 		LatchWaitSet = NULL;
 	}
 
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 	close(selfpipe_readfd);
 	close(selfpipe_writefd);
 	selfpipe_readfd = -1;
 	selfpipe_writefd = -1;
 	selfpipe_owner_pid = InvalidPid;
+#endif
+
+#if defined(WAIT_USE_EPOLL)
+	close(signal_fd);
+	signal_fd = -1;
+#endif
 }
 
 /*
@@ -299,10 +335,10 @@ InitLatch(Latch *latch)
 	latch->owner_pid = MyProcPid;
 	latch->is_shared = false;
 
-#ifndef WIN32
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 	/* Assert InitializeLatchSupport has been called in this process */
 	Assert(selfpipe_readfd >= 0 && selfpipe_owner_pid == MyProcPid);
-#else
+#elif defined(WAIT_USE_WIN32)
 	latch->event = CreateEvent(NULL, TRUE, FALSE, NULL);
 	if (latch->event == NULL)
 		elog(ERROR, "CreateEvent failed: error code %lu", GetLastError());
@@ -363,7 +399,7 @@ OwnLatch(Latch *latch)
 	/* Sanity checks */
 	Assert(latch->is_shared);
 
-#ifndef WIN32
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 	/* Assert InitializeLatchSupport has been called in this process */
 	Assert(selfpipe_readfd >= 0 && selfpipe_owner_pid == MyProcPid);
 #endif
@@ -550,9 +586,9 @@ SetLatch(Latch *latch)
 
 	/*
 	 * See if anyone's waiting for the latch. It can be the current process if
-	 * we're in a signal handler. We use the self-pipe to wake up the
-	 * poll()/epoll_wait() in that case. If it's another process, send a
-	 * signal.
+	 * we're in a signal handler. We use the self-pipe or SIGURG to ourselves
+	 * to wake up WaitEventSetWaitBlock() without races in that case. If it's
+	 * another process, send a signal.
 	 *
 	 * Fetch owner_pid only once, in case the latch is concurrently getting
 	 * owned or disowned. XXX: This assumes that pid_t is atomic, which isn't
@@ -575,11 +611,17 @@ SetLatch(Latch *latch)
 		return;
 	else if (owner_pid == MyProcPid)
 	{
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 		if (waiting)
 			sendSelfPipeByte();
+#else
+		if (waiting)
+			kill(MyProcPid, SIGURG);
+#endif
 	}
 	else
 		kill(owner_pid, SIGURG);
+
 #else
 
 	/*
@@ -856,8 +898,13 @@ AddWaitEventToSet(WaitEventSet *set, uint32 events, pgsocket fd, Latch *latch,
 	{
 		set->latch = latch;
 		set->latch_pos = event->pos;
-#ifndef WIN32
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
 		event->fd = selfpipe_readfd;
+#elif defined(WAIT_USE_EPOLL)
+		event->fd = signal_fd;
+#else
+		event->fd = PGINVALID_SOCKET;
+		return event->pos;
 #endif
 	}
 	else if (events == WL_POSTMASTER_DEATH)
@@ -932,12 +979,13 @@ ModifyWaitEvent(WaitEventSet *set, int pos, uint32 events, Latch *latch)
 		if (latch && latch->owner_pid != MyProcPid)
 			elog(ERROR, "cannot wait on a latch owned by another process");
 		set->latch = latch;
+
 		/*
 		 * On Unix, we don't need to modify the kernel object because the
-		 * underlying pipe is the same for all latches so we can return
-		 * immediately.  On Windows, we need to update our array of handles,
-		 * but we leave the old one in place and tolerate spurious wakeups if
-		 * the latch is disabled.
+		 * underlying pipe (if there is one) is the same for all latches so we
+		 * can return immediately.  On Windows, we need to update our array of
+		 * handles, but we leave the old one in place and tolerate spurious
+		 * wakeups if the latch is disabled.
 		 */
 #if defined(WAIT_USE_WIN32)
 		if (!latch)
@@ -1421,8 +1469,8 @@ WaitEventSetWaitBlock(WaitEventSet *set, int cur_timeout,
 		if (cur_event->events == WL_LATCH_SET &&
 			cur_epoll_event->events & (EPOLLIN | EPOLLERR | EPOLLHUP))
 		{
-			/* There's data in the self-pipe, clear it. */
-			drainSelfPipe();
+			/* Drain the signalfd. */
+			drain();
 
 			if (set->latch && set->latch->is_set)
 			{
@@ -1575,7 +1623,7 @@ WaitEventSetWaitBlock(WaitEventSet *set, int cur_timeout,
 			cur_kqueue_event->filter == EVFILT_READ)
 		{
 			/* There's data in the self-pipe, clear it. */
-			drainSelfPipe();
+			drain();
 
 			if (set->latch && set->latch->is_set)
 			{
@@ -1691,7 +1739,7 @@ WaitEventSetWaitBlock(WaitEventSet *set, int cur_timeout,
 			(cur_pollfd->revents & (POLLIN | POLLHUP | POLLERR | POLLNVAL)))
 		{
 			/* There's data in the self-pipe, clear it. */
-			drainSelfPipe();
+			drain();
 
 			if (set->latch && set->latch->is_set)
 			{
@@ -1951,7 +1999,8 @@ WaitEventSetWaitBlock(WaitEventSet *set, int cur_timeout,
 }
 #endif
 
-#ifndef WIN32
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_KQUEUE)
+
 /*
  * SetLatch uses SIGURG to wake up the process waiting on the latch.
  *
@@ -1967,10 +2016,8 @@ latch_sigurg_handler(SIGNAL_ARGS)
 
 	errno = save_errno;
 }
-#endif							/* !WIN32 */
 
 /* Send one byte to the self-pipe, to wake up WaitLatch */
-#ifndef WIN32
 static void
 sendSelfPipeByte(void)
 {
@@ -2000,45 +2047,58 @@ sendSelfPipeByte(void)
 		return;
 	}
 }
-#endif							/* !WIN32 */
+
+#endif
+
+#if defined(WAIT_USE_POLL) || defined(WAIT_USE_EPOLL)
 
 /*
- * Read all available data from the self-pipe
+ * Read all available data from self-pipe or signalfd.
  *
  * Note: this is only called when waiting = true.  If it fails and doesn't
  * return, it must reset that flag first (though ideally, this will never
  * happen).
  */
-#ifndef WIN32
 static void
-drainSelfPipe(void)
+drain(void)
 {
-	/*
-	 * There shouldn't normally be more than one byte in the pipe, or maybe a
-	 * few bytes if multiple processes run SetLatch at the same instant.
-	 */
-	char		buf[16];
+	char		buf[1024];
 	int			rc;
+	int			fd;
+
+#ifdef WAIT_USE_POLL
+	fd = selfpipe_readfd;
+#else
+	fd = signal_fd;
+#endif
 
 	for (;;)
 	{
-		rc = read(selfpipe_readfd, buf, sizeof(buf));
+		rc = read(fd, buf, sizeof(buf));
 		if (rc < 0)
 		{
 			if (errno == EAGAIN || errno == EWOULDBLOCK)
-				break;			/* the pipe is empty */
+				break;			/* the descriptor is empty */
 			else if (errno == EINTR)
 				continue;		/* retry */
 			else
 			{
 				waiting = false;
+#ifdef WAIT_USE_POLL
 				elog(ERROR, "read() on self-pipe failed: %m");
+#else
+				elog(ERROR, "read() on signalfd failed: %m");
+#endif
 			}
 		}
 		else if (rc == 0)
 		{
 			waiting = false;
+#ifdef WAIT_USE_POLL
 			elog(ERROR, "unexpected EOF on self-pipe");
+#else
+			elog(ERROR, "unexpected EOF on signalfd");
+#endif
 		}
 		else if (rc < sizeof(buf))
 		{
@@ -2048,4 +2108,5 @@ drainSelfPipe(void)
 		/* else buffer wasn't big enough, so read again */
 	}
 }
-#endif							/* !WIN32 */
+
+#endif
diff --git a/src/backend/utils/init/miscinit.c b/src/backend/utils/init/miscinit.c
@@ -118,6 +118,11 @@ InitPostmasterChild(void)
 	/* We don't want the postmaster's proc_exit() handlers */
 	on_exit_reset();
 
+	/* In EXEC_BACKEND case we will not have inherited BlockSig etc values */
+#ifdef EXEC_BACKEND
+	pqinitmask();
+#endif
+
 	/* Initialize process-local latch support */
 	InitializeLatchSupport();
 	MyLatch = &LocalLatchData;
@@ -135,11 +140,6 @@ InitPostmasterChild(void)
 		elog(FATAL, "setsid() failed: %m");
 #endif
 
-	/* In EXEC_BACKEND case we will not have inherited BlockSig etc values */
-#ifdef EXEC_BACKEND
-	pqinitmask();
-#endif
-
 	/*
 	 * Every postmaster child process is expected to respond promptly to
 	 * SIGQUIT at all times.  Therefore we centrally remove SIGQUIT from
